Spatiotemporally structured noise, such as physiological noise, is a potential source of artifacts in functional magnetic resonance imaging (fMRI) and is the main limiting factor for the detection of small blood oxygen level-dependent (BOLD) signal variations. fMRI was employed to detect low-frequency BOLD signal fluctuations, which are thought to be related to spontaneous neuronal activity in the resting human brain. The sensitivity to noise, that is, signal variations of non-BOLD origin, was investigated for two- (2D) and three-dimensional (3D) imaging techniques. Incomplete relaxation between subsequent scans increases the level of temporally and spatially correlated signal variations originating from physiological and/or systemic noise. Although inflow effects are suspected to be reduced in 3D echo-planar imaging (EPI) compared with multi-slice 2D EPI, the noise level was higher in the 3D technique. The noise level in 3D fMRI experiments was significantly increased by instabilities of the transverse steady-state magnetization as the repetition time was of the order of T(2). By implementing radiofrequency spoiling, temporal signal fluctuations and erroneous inter-regional correlation in connectivity maps were diminished to a level present in data sets acquired with 2D EPI.